fix(ci): the PR gates accept the readable rationale and let the body override a stale branch

[mrveiss]

Closes #16050, #16051.

Thinking Path

Two gates in .github/workflows/pr-issue-validation.yml failed PRs today for reasons that were correct about their own rule and wrong about their premise. Both are in one workflow and one script, so they land together as one review pass and one CI run.

#16050 — the batching gate rejects the form people write

check_pr_issue_batching.py required Single-issue rationale: with the reason on the same line. A markdown heading with the reason beneath matched nothing.

Two independent sessions hit it within four hours — #16018, and #16049 which is the CRITICAL auth fix for #16040. Two authors independently choosing a form the checker rejects is the checker's defect, and the rejected form is the better one: it matches the section style the rest of the body already uses (## Thinking Path, ## What Changed, ## Verification).

The hint taught the failing shape. It printed the inline form indented, as an example, which reads as illustration rather than exact-match.

#16051 — the issue-link gate trusts a name that cannot be corrected

The gate derives the expected issue from the branch name. Two sessions reserved number 16040 six minutes apart; the loser's branch was already issue-16040-*, and the number went to a CRITICAL auth issue. They rewrote their commits to the right number — and the gate then failed the PR for not referencing the wrong one, instructing them in its own error message to add it back.

A gate whose remediation re-creates the defect is worse than no gate. And there is no in-place fix: GitHub does not allow changing a PR's head branch.

The decision was already made — #14241 accepts the body over the branch, and scoped it to forks on the reasoning that an outside contributor cannot be asked to rename. An internal author cannot either. The fork restriction drew the line at who could rename; the answer is nobody.

What Changed

• check_pr_issue_batching.py accepts ## Single-issue rationale with prose beneath, scanning past blank lines and stopping at the next heading. The hint now names both forms.
• pr-issue-validation.yml drops PR_IS_FORK from the body-over-branch escape, and its failure message now calls the branch-derived number a guess, telling the author an explicit Closes #<real issue> overrides it.

Verification

31 tests pass. The two that matter are contrasts, because both changes are widenings and the cheap wrong version of each is to widen too far:

case	result
heading with prose beneath	accepted
reason several blank lines down	accepted
heading with nothing under it	rejected
heading immediately followed by ## What Changed	rejected — does not borrow the next section
heading at end of body	rejected
inline form, blank after colon	still rejected

The empty-section case is the load-bearing one. This gate asks whether a human justified standing alone; a fix that matches the heading and stops looking turns it into a keyword check that any PR passes with an empty heading.

The escape's regex was exercised directly:

body	escape fires?
Closes #16046, #16044 (the #16047 case)	yes
Refs #16040 only	no — not a closing claim
links nothing	no
this self-closes #123	no

So it widens who may override the inference, not what counts as one. A PR linking no issue at all still fails.

Model Used

Opus 5

[coderabbitai]

Important

• 🔍 Trigger review

This repository does not receive automatic reviews because it has fewer than 10 stars.

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: ASSERTIVE

Plan: Advanced

Run ID: c4451ef1-625c-4cb9-93fa-ef4f9031b4d4

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[github-actions]

✅ SSOT Configuration Compliance: Passing

🎉 No new hardcoded values of either class — ssot and other both block.

Known backlog in pipeline-scripts/hardcoded_values_baseline.txt is suppressed and tracked in #14371.